Kafka vs Amazon SQS

At the morning peak, the bike-share service publishes 1,200 events a second, and four readers need every one of them. On Amazon SQS, each reader gets its own queue, so 1,200 events become 4,800 stored messages a second, each received and deleted separately. Standard queues support "a nearly unlimited number of API calls per second," so the rate alone is not a problem.

The harder number is 30. Forecasting retrains on the last 30 days of rides, and an SQS queue keeps a message for at most 14 days